AWS is Dead. New Oracle + Nvidia Datacenters Win In Early KO
Wow. It has been an absolutely wild ride since we last spoke about datacenters. In that post we talked about the possible emergence of a new global datacenter ecosystem as new AI and AR/VR workloads provide an opening in the Infrastructure-as-a-Service industry (what AWS is). AWS has dominated since the start of the last bubble. We speculated that, perhaps, there was a chance here for a change in the market. One that we thought, most likely, would come from the seemingly unlikely and yet totally natural affinity between two of our tech giants, Nvidia and Oracle.
I am totally thrilled to share that recent announcements about the Nvidia + Oracle collab, have just in the past week confirmed they are building out a global network of small datacenters that are built from the atom up by Nvidia and Oracle to handle the new kind of technology development that will be done in this era and to handle the new application patterns of AI apps and AR/VR environments, the massive increase in data collection, and the new generations of consumer and enterprise hardware that are coming.
We have new datacenters, folks. This is it. This is the thing. This is a warp. Nvidia and Oracle built them for us, from an atomic level, from top to bottom, for the exact ubiquitous computing moment that we are entering.
This is already over. AWS is fucking dead. Anything you see from here on out is blood and gargling.
By the time that the majority of the industry was talking about ec2 in the last bubble, it was already over. The paradigm shift in management of compute power was monumental and it immediately became the fundamental platform for most of what you experienced of web 2.0. We have just seen another paradigm shift here.
Nvidia and Oracle teaming up to build AI supercomputer datacenters with the goal of exposing the full capabilities of the datacenter itself to the application. As Jensen said, the computer is the datacenter.
Yep. Done.
This is change in the development model, it’s a change in how we build fucking datacenters, it is a change in the application paradigm, it is a change in the collaborative model, it is a new formation to address the new latency requirements, the new performance requirements, the processing requirements. It is a change in how we bring technology to the world. Top to bottom.
What has happened here is that Nvidia and Oracle are the new IaaS platform, they will dominate the infrastructure of this generation of technology, they are AWS now, and AWS is now mortally wounded, the way Rackspace once was. The bleeding will begin shortly, but it’s already over. This paradigm shift is much more up these folks alley than Amazon, on multiple areas which we will touch on in this post. In this area, Oracle and Nvidia dominate.
There is no hope that Amazon will compete. The production of infrastructure at this level is done from extremely deep parts in the corporate anatomy and in a very fundamental way, you either have what that moment needs with the capabilities you have or you don’t. This could not have come from any other individual companies or the connections between them. There is no way to fake infrastructure like this.
This is no disrespect to AWS, except for some collegial ribbing. They have done great work and I share Larry’s abiding respect for your engineering teams. You have all served us very well in web 2.0 and we would not be here without you. That said, we have new global infrastructure now that was custom built with the best minds in physics and artificial intelligence, by people who work at a level of latency that is implicated here and have for a very long time. Amazon was cheap commodity boxes and co-tenancy. Now we have very expensive boxes and we have arrived at a new model for how to deliver these supercomputing capacities. When you have these extreme performance requirements, every application and every possible part of the stack must be extremely tight to get the level of performance and speed to deliver these new kinds of abilities and capabilities. Every.single.thing must sing. This is what Jensen is talking about when he talks about a “full stack problem”. That is the main differentiator here. The bringing to bear of the entire stack from top to bottom — the whole fucking datacenter, all of it — in a precision way to the new world.
Correct, I am saying that we have a new king and the victory is already definitive. The infrastructure race is decided very early in the game because that is the substrate that the rest of the bubble uses to grow: compute, data, and money are substrate. What is happening here is that these companies are readying to release, to the entire technology economy of waiting startups and enterprises, a world of compute power and speed that blows what we have been dealing with out of the water, that is going to put a new level of computing into production, that aims to make that level of computing ubiquitous, and to make a new level of computing available to the entire technology field and what it brings in.
Without the gameplay from the datacenters, the bubble can’t happen. This is a precondition for the bubble and it is decided early. Getting a critical mass of development on top of infrastructure is also what allows us to expand the capabilities, refine the technology, get network effects, and consolidate data. So it is not only important that this exists but also that we get as much shit as possible onto it. This is power law. We push forward together and this is an area where you see the increasing level of consolidation that are needed to support the technology ecosystem performing at this level.
The first big clue for me was when Larry Ellison made a rare and celebrated appearance at an Oracle analyst event in the end of March, where he craftily made some mega-announcements: 1.) We are building a new, globally distributed datacenter infrastructure that is fundamentally different from what is on the market today 2.) We actually built the fastest computer in the world. 3.) You can get the fastest computers in the world from my new datacenters (#3, implied).
Larry Ellison is not a man who fucks around. Unlike some our elder statesmen like Marc Andreessen who do absolutely nothing BUT fuck around; while Marc has been fanning the flames of increasingly more hellish FUD as some kind of aggro incel marketing exercise for OpenAI, Ellison was doing things like casually mentioning the biggest development in datacenters since ec2…
which launched in 2006.
It’s over lmfao.
This came in the wake of announcements about a massive collaboration between Nvidia and Oracle on AI, VR/AR workloads and machines. I was absolutely frothing with excitement. Then in the last few days, Nvidia released an earnings report and hit the legendary 1 trillion market cap!!!! Wow. And I’m telling you that that is some of the most solid money in the business because Nvidia, which specializes in PHYSICS, it is a PHYSICS COMPANY, is basically our bleeding edge of what we are fundamentally able to achieve in compute. For this reason Nvidia is universally respected within the industry. Unless physics goes broke, Nvidia will not.
And the CEO of Nvidia, Jensen Huang, gave his first live presentation, in fucking four years. Now, that is the thing about Nvidia: they work in silence. People think Apple moves in silence, but let’s face it, they are kinda noisy and new money with it. Nvidia operates in a silence not because of elaborate psyops (there probably are), but in large part because 1.) no one understands wtf they do, they are our bleeding edge, how do you fuck with what you cannot comprehend and. 2.) Nvidia is universally respected in the industry because we rely on their achievements in physics to shape the entire frame of what is possible. No one wants to fuck with them. We let them work.
The other reason this is a knockout is because this is the solution they have been forced to come to by the needs they have as a company. Specifically, they cannot get what they need out of the machines without having a stack that is built top to bottom, using full stack in one of the broadest meanings of the term ever used, by them. They need the wholeeeee datacenter to be done top to bottom for this. They can’t get what THEY want unless they do this.
Among others companies, this is THEIR existential problem and that is why they will win. In this talk, Jensen announced something really fucking spectacular, which was: datacenters. Datacenters built from top to bottom, soup to nuts, by Nvidia. We are talking about, the physics company is building the entire datacenter now. And oh yeah. With Oracle.
What we are seeing in the emergence of these datacenters is an existential solution. That is why the game is done.
You heard it now folks, two times from both parties: the Nvidia and Oracle GLOBAL AI SUPERCOMPUTER DATACENTERS have arrived.
It sounds like winning because it is.
That cannot be competed with. Let me be among the first to congratulate both Larry Ellison and Jensen Huang on a decisive victory. You are both guardians of the edge, and I wish you all the luck in the world.
This means we have our web 3.0 AWS. And AWS was always just Oracle with clever networking. So where have actually arrived is back to what made the first bubble POSSIBLE on anyyyy level: Oracle.
We’re back folks.
Again, Nvidia is a physics company and ultimately they are building physics as a service in a much realer way than anything we’ve seen before. So they are building out a global network of BRAND NEW, finely tuned, data centers that are physics-level-first, developed specifically for this new world of AI apps and platforms and virtual reality of many kinds, for the new data loads, for the new hardware plays. When Jensen says that he is looking at the datacenter as a “full stack problem” you can see the glimpse of respect for THE PROBLEM SPACE which is so deeply held and the mark of a true tech executive.
When he says “the computer is the datacenter”, he is talking about a paradigm shift in fucking development itself which is exact what AWS posed at the time. This is the real deal. AWS is toast. They will sell AWS datacenters like luxury yurts on Airbnb. Larry will be the super host. And then he’ll turn it into a Four Seasons. Complete with a yacht expedition on a boat made entirely from salvaged parts from the US East availability zone.
There are a lot of things that change with these workloads and that is the reason that AWS is already dead. Most of their shit — from the machines to the engineers to the DNA to the legacy revenue and data, to their datacenters themselves, is for the old workloads. These new datacenters are purpose-built, with no bifurcation in focus, for this new world of apps wheres the vast majority of Amazon infrastructure is not set up for this — in fact, web 2.0 is hosted in majority inside of AWS. AKA the old web.
Their entire business model is wrapped around a web 2.0 data volume and speed and a web 2.0 consumer platform world, a totally different application model and platform model and different underlying technologies. The stack for a web 2.0 iPhone app of ANY scale is going to look totally different than one for cutting-edge applications using AI, AR and VR, and the new form factors that are launching, etc. In the tech industry, we are all told, and tell each other: latency will kill you. The statistics on this are insane; it’s something like by the time you reach a fragment in time to load first page you are gonna lose like, basically every single customer. That is the application pattern that was developed in the last bubble, and the latency requirements for those applications are different at every single level — full stack — from the those of this new world. I.e., latency needs to go down.
Way down.
You know what Nvidia does? Hunts latency at levels most AWS engineers don’t think in.
I feel like the hints that these are going to be hundreds of small datacenters has something to do with latency.
Web 2.0 is being officially deprecated. People have been too busy making fun of the “Web 3.0” concept, utterly clueless that this is actually a technical delineation and one that was not arrived at by accident for use here. In computer science major releases are marked by whole numbers. That means that this is a major release from the whole industry. What you are about to see in the world around you in some ways does come from all of us. This was all planned.
And for Amazon, well, most data loses value very quickly across time. Real time data and the data that starts emitting from new technology, quickly supplants it especially as this data will be much more expansive in all meanings of the word. The Web 2.0 infrastructure and apps and companies will disappear. Glacial wipeout. Clean slate. The value of data goes down very quickly and suddenly. AWS is stuffed to the gills with web 2.0 applications that soon will not being maintained at all. The value of their assets — the data — takes a massive depreciation when it is no longer “hot data” or data that is being served to a client of some form at high concurrency. This data loses value quickly as new kinds and levels of data are reached. One of the important aspects of the transition we are making is what level of capacity, performance, etc. ,Is that we are going to need to support huge influxes of novel data that is already being collected by AI and by new form factors and by new application and startups that have reached an entirely new volume of data.
Volume, yes. But the real problem… is the speed. Again, this is where Amazon simply can not compete. They have not been dealing with this problem on this level. Nvidia is literally where the core, core computing technology for immersive systems and immersive computing environments, has come from. That is their DNA. That is who they are. They are the closest of any other giant to supporting these types of experiences. And that goes into and is reflected at every single level of the company and what it produces. That is why these things are a lot less mysterious than they seem, when you understand the nature of how deeply related the company is to what it is able to achieve or not achieve or compete or not compete with.
As Jensen spoke about in his recent talk, AI supercomputing will literally be in everything around us. Applications will be wrapped around AI the way they once wrapped around the raw data. The AI compute is the platform. A world of ubiquitous AI computing is the story that Jensen was telling us, and these are the parties that can make this a possibility. Without the infrastructure for AI, there is no AI. So again, just great work again here everyone.
I think that is the vision that Jensen has, and his proposal for how to manage this and what design proprieties we are looking at, an engineering plan that unfolds over time.
They literally mean a new internet. Think about it. We’re getting fucking new datacenters. Honestly, that’s all you need to know. That is at the base of bubbles, it is the innovation in datacenters and what is inside them, that provides the entire technical ecology that a bubble rests along. This is why it is so lucrative and a position of power.
This move is brilliant not only because of the technical DNA here, though that is significant, but one place where it really sings is the fact that Oracle is the gold standard in enterprise sales. I.e., the art of forcing companies to adopt and adjust to new technologies.
One thing that many people don’t know about the technology industry is that, companies do not come coming to us bursting with new ideas about how to make their technology department better. Of course they WANT, theoretically, to be doing that, but what is actually done, is, what actually is done is we send massive amounts of highly trained people into all of the corporations and we INITIATE a process of forcefully integrating with them.
Something that is very tedious and exhausting because these companies STILL do not know how to work with technologies and the industry, do not know how to maintain in-house competency, and live in another world all day where the focus on their core mission, not computers. While there are some large enterprises that have very respectable computing teams, those have also been carefully nurtured by the technology vendors themselves and they are formed in our image; satellites, if you will. Every major enterprise installation is, for the most part, manufactured, sold and installed by us. We come up with how you’re gonna use it, how we’re gonna integrate it, we will design you a business plan, we will hire you a new team, we will send you one of our teams, we will be your marketing department, we will do whatever. Come to our conferences. Let’s bring a meetup to you. Maybe we can get you signed up for this beta. What if we threw in the hardware for you?
We sell computers here.
At the end of the day, we sell computers. We are the computer department. We bring them to you and we figure out how to hook them puppies up to what you are doing. We come in and install it for you and most of the time we run the vast majority of that for you as well. Even when there is big demand and anxiety about keeping up with the newest trends, we are the computer guys.
That means web 3.0 is just as much of a technical issue as it is, going around knocking on doors. Some of these people have not had new computing capabilities since the last time Larry Ellison showed up at their door. There are truly people who will be getting leapfrogged from Oracle to Oracle without ever stopping at AWS. Perhaps these are the real winners.
The point being, we can have magical ubiquitous AI supercompute coming out of our asses but we are still going to need to personally install this into a huge range of large scale environments in the real world and
that my friends
is where Oracle absolutely shines. Absolute legend in the enterprise sales department. Look, Amazon has absolutely done a lot of white gloving and hell, they even built the CIA its own datacenters. But, they are not an enterprise sales company. They are not our best foot forward and THAT is what Oracle does best and what will be sorely needed here. Someone needs to go around, basically create an entire platform for these companies in the AI world, a bunch of apps, lots of new shit, every stop needs to be totally re-outfitted, and that is where Oracle takes the cake. Showing up at your doorstep. With a datacenter for you to buy. So you not only have that combination of the technical chops but you have the powerhouse that did the FIRST , never can be out-done, global compute infrastructure, on hand for this one. They installed technology into the entire world by hand for the first bubble. And what we are looking at here looks more like THAT than AWS in Web 2.0, tbh.
Amazon basically took Oracle and hooked up the internet to it. It’s smart, it was good, it’s clever, but the thing Amazon was releasing was just a better way to do things we knew how to do. When Oracle first went in, and when Oracle AI goes in now, that is an actually new computing paradigm. And I bet they have a lot more surprises for us. And I think for widespread, ubiquitous AI supercomputing, Oracle plays a massive role in getting the entire world — not just tech, not at all, everyone — up onto this shit.
One of the things I really like about this partnership is that, look, of course Nvidia and Oracle are bad companies in the sense that these are capitalist companies and so of course the general fare of capitalism. However, I am not indiscriminate. I have been watching these companies operate for a long time, and they are simply not as bad as many of our players on the board. This is not necessarily through the innate character of the CEOs - though to exist, it in some way inevitably is — but Oracle and Nvidia have conducted themselves with class and professionalism when at all points the rest of the Valley has been pissing, throwing up, starting fascist political parties, distasteful marketing stunts, and nurturing hate groups on fucking 8chan, lots of getting wasted and sexual abuse, just really ugly stuff. While I don’t think Larry Ellison is gonna get a straight up pass into Heaven necessarily, that kinda comes with the billionaire thing, there is absolutely a “scale of evil” going on here, and I would much rather they have this play than other parties I could think of. And I will be putting in a word with heaven for Larry.
The motivation at the infrastructure layer is to hold up the computing world and responding to those technical demands, so you essentially have an obligation to the entire global computing infrastructure and your revenues are tied to that. That is very different than say, a venture capitalist. That said, these companies are part of an industry which has had some pretty devastating results. Larry and Jensen remember a San Francisco before we moved in McMansion apartment complexes (which I personally love) and basically just took the whole damn thing over. That’s not fucking woke to say, grow up. That’s a material thing that happened. We all saw it happen. The tech bubble has had terrible economic effects on these areas.
However, in Oracle and Nvidia, I don’t see little meme armies, fake ass media publications, pathological lying, fascist propaganda, sick games, open sadism, twisted bullshit, war hawking, and the just total open debauchery that we see in trash like a16z. Like when Marc Andreessen comes on stage you are about to get a full assault on reason by a master troll. When Ellison or Huang speak, they are going to be talking to you about fucking physics and databases, okay. Everytime. Like Nvidia does not have fucking time. They are in the fucking future. Every second that Jensen experiences is like some extremely lengthy period of another dimension he is either in or observing. No one knows. We do not know.
I’m sure the absolute furthest that Nvidia has gotten would blow all our fucking brains. Like I’m all worried about Anduril blowing us all up with a nuclear bomb while Marc Andreessen chills in space waiting for the grass to come back on his new planet. But honestly our most likely existential threat is not AIG but rather, a “lab spill” over at Nvidia that just annihilates the entire plane of existence. It is basically unethical to waste even a single second of Jensen’s time and that’s basically that. He is reaching into the time box and all of his energy must be focused on making sure we all make it out of whatever comes back.
Jensen deals in the world of existential threat at all times and he has lived in that world for a very long time. It is this extreme computing that AWS is NOT known for, that is not where their bread and butter is, they haven’t been having to deal with these kinds of issues on this level. And Nvidia has been at the forefront of designing in this new world while AWS has been hooking up wifi to a cheap box of commodity hardware and passing out mobile apps. That infrastructure is now literally 20 years old. Their major database buildout is way outdated. And the data they got in web 2.0 is a lot of unrealized losses.
We are looking at a whole never level of datacenter here folks. Jensen was talking about the datacenter as a full-stack problem, and honestly, shivers went down my back and hairs stood up on the back of my neck. His “full stack” is not what most people’s full stack is. Not at all.
For Jensen, everything above the atom is the application layer.
When we talk about existential threats from computers, it won’t come out of fucking OpenAI. They do not build or maintain their own compute infrastructure. That is why they are with Microsoft. They are at a wayyyyy higher degree of abstraction than Nvidia or Oracle is at. If a power cord needs to be pulled out of a datacenter or a fucking availability zone needs to be cut off because something has gone rogue, we will probably have to drone bomb and Nvidia and Oracle datacenters. However, good the fuck luck taking one of their systems off-line. It will also be all built with web 3.0 security systems so we will definitely be seeing paramilitarized datacenter defense, where we are the ones operating that defense; not the military per se.
Jensen is probably the person who ACTUALLY thinks the most about mass death scenarios. Millions of ways to die by computer. Because of how far out Nvidia works at the edge, this terrain is one they have navigated for a long time. Their much more considered approach reveals the subject matter RESPECT that is required. What a16z has been saying has no respect for even the subject at hand. If a16z could do shit with physics the way Nvidia does, we would have all died right after Steve Jobs announced the iPhone.
The fact we are alive is in fact, a testament, in some twisted way, to these folks.
Jensen’s vision is the best we have on the playing field right now for where these technologies could go. Jensen has some very beautiful and subtle language for dealing with AI and for thinking about AI. This thinking is far more engaging, very plush, generative, curious and sensitive, then we have seen in other places. We see OpenAI, funded by a16z, out in the marketplace trying to make sure we are afraid that we’re all gonna die from this. I’m not totally clear on their motives there and they are probably manifold, something we’ll explore in a subsequent post. But this reflects the violent heart of the VCs who back OpenAI, and another example where you are seeing serious market manipulation and psyops coming out of these people.
Jensen, in contrast, spoke about “infusing” experiences and “animating” them with AI. This is a world where AI isn’t trying to fool you into thinking it is a person, and where it isn’t trying to kill you, but rather, a world where AI is soft, beautifying, enhancing, illuminating. Though working at the most elemental level of these machines, this vision is humanistic, not degrading of humanity and its limitations. The vision of OpenAI is AI as a blunt-force hyperspeed crazy train to nowhere, terrifying, Terminator-esque. That is venture capital. The first thing they put out tries to copy and imitate a human being. Trolling. This is not the approach of Nvidia.
Nvidia is an extremely precise company and this language emerges from the technical properties of what THEY are building and THEIR vision for where it goes. This was not just marketing language, I promise you that this is language used at all levels of the company and words that inform all of their work. They showed a respect for humanity, for human autonomy and for respect of the natural boundaries of humanity, that has been utterly absent in Meta’s conception, and equally absent in a16z/OpenAI’s perspective.
Responsible stewardship of technology is the bulk of our responsibility if not of our action. And each technology executive has a different view of what that looks like. Ironic that the lowest down in the stack, down to atoms, does a better job of interfacing with humans that people who work on the application or sociopath layer. Maybe because Nvidia is not run by psychopaths and sociopaths. A hard thing to imagine in tech, but it’s possible that that is what we are seeing here. Wow.
This is a much more powerful, compelling, gentle, framework for thinking about AI. And it is for that reason I must believe they will win. The language we use absolutely impacts the whole stack and the language is abso fucking luetly, a part of that stack and was built into that stack by some very serious and thoughtful humans. This is humans working on AI. That is why they will win.